Create skills from scratch. Port skills from anywhere.
You spent 45 minutes perfecting a complex multi-step process in Amplifier. Next week you'll need it again — but the knowledge lives only in that session.
You have battle-tested skills from Claude Code or Cursor, but they use different tool names, paths, and frontmatter conventions.
Until now, you had to manually author SKILL.md files from scratch — knowing the full Agent Skills spec, Amplifier extensions, and conventions scattered across multiple docs.
load_skill() to catch issues before you commit.Invoke with: /skillify or just say "create a skill" / "save this as a skill"
Translates skills from Claude Code, Cursor, or any AI coding assistant into Amplifier-native format.
Tool names (Read → read_file), paths (.claude/ → .amplifier/), frontmatter fields — all handled.
Encounters an unfamiliar platform convention? It researches the source platform's docs before guessing.
Spawns a delegate self-session to exercise the adapted skill — verifying it actually works in Amplifier.
8-step workflow including platform research, mapping, writing, and a full testing step.
Invoke with: /adapt-skill path/to/source.md or "port this skill to Amplifier"
We brainstormed the design section by section, iterating until the skill worked end to end.
We walked through porting a Claude Code skill to Amplifier, working out the mapping rules live.
We literally used /skillify to capture the adaptation workflow we'd just performed.
That capture became /adapt-skill — the first skill ever created using the skillify skill itself.
It's skills all the way down. 🐢
load_skill("skills-assist")skills-assist now includes 5 companion files — including a brand new testing-guide.md for verifying skills before deployment.
Don't dump everything at once, don't micro-ask one question per turn. Group related decisions into natural clusters.
Trigger phrases belong in the description field — that's what the visibility hook shows. The body loads after invocation. Too late.
Forked skills run as one-shot subagents. If your skill needs multi-turn conversation with the user, it must be inline.
Both skills include a dedicated testing step using load_skill() and optionally delegate(agent="self").
Available now in the latest amplifier-bundle-skills.
Feedback and PRs welcome.